Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Persist context across engineering sessions — save important decisions, gotchas, or patterns that should survive beyond the current conversation. For long-term workspace memory, prefer mempalace.
.claude/skills/evolution-foundation-dev-remember/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -13% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 406% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 15% | 0% |
Derived from oh-my-claudecode (MIT, Yeachan Heo). Adapted for the EvoNexus Engineering Layer.
Quick context persistence for engineering sessions. Save decisions, gotchas, or patterns that should survive the current conversation but don't deserve a full memory entry in mempalace.
.claude/agent-memory/{agent}/ insteadmemory/ (mempalace owns this)Engineering layer notes go to the relevant agent's memory folder:
.claude/agent-memory/apex-architect/.claude/agent-memory/hawk-debugger/.claude/agent-memory/grid-tester/If the note doesn't fit any single agent, save to workspace/development/research/[C]remember-{topic}-{date}.md.
Each note has:
For long-term, cross-session, cross-agent shared memory, prefer mempalace:
dev-remember is for engineering-specific quick saves. mempalace is for everything else.
prod-memory-management (the Clawdia skill that owns shared memory)| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 7,584 | 5,164 | -32% | 1 | 1 | 0% | 1,362 | 1,498 | +10% | 0 | 0 | — |
case-02 | fail→pass | 9,575 | 4,920 | -49% | 1 | 1 | 0% | 1,494 | 1,346 | -10% | 0 | 0 | — |
case-03 | fail→pass | 8,614 | 9,290 | +8% | 1 | 1 | 0% | 1,417 | 1,237 | -13% | 0 | 0 | — |
case-04 | fail→pass | 1,938 | 6,233 | +222% | 1 | 1 | 0% | 298 | 1,509 | +406% | 0 | 0 | — |
case-05 | fail→pass | 6,581 | 3,965 | -40% | 1 | 1 | 0% | 1,089 | 1,253 | +15% | 0 | 0 | — |
case-06 | fail→pass | 7,737 | 4,258 | -45% | 1 | 1 | 0% | 1,261 | 1,265 | +0% | 0 | 0 | — |
case-07 | fail→pass | 9,083 | 3,600 | -60% | 1 | 1 | 0% | 1,723 | 1,223 | -29% | 0 | 0 | — |
case-08 | fail→pass | 10,967 | 3,864 | -65% | 1 | 1 | 0% | 1,719 | 1,167 | -32% | 0 | 0 | — |
case-09 | fail→pass | 10,841 | 4,358 | -60% | 1 | 1 | 0% | 1,792 | 1,192 | -33% | 0 | 0 | — |
case-10 | pass→pass | 5,799 | 3,040 | -48% | 1 | 1 | 0% | 829 | 991 | +20% | 0 | 0 | — |
case-11 | fail→fail | 6,918 | 3,735 | -46% | 1 | 1 | 0% | 1,211 | 1,099 | -9% | 0 | 0 | — |
case-12 | pass→pass | 8,183 | 5,022 | -39% | 1 | 1 | 0% | 1,300 | 1,333 | +3% | 0 | 0 | — |
case-13 | pass→pass | 6,274 | 2,803 | -55% | 1 | 1 | 0% | 1,042 | 1,011 | -3% | 0 | 0 | — |
case-14 | pass→pass | 8,653 | 4,606 | -47% | 1 | 1 | 0% | 1,480 | 1,262 | -15% | 0 | 0 | — |
case-15 | fail→pass | 11,075 | 5,187 | -53% | 1 | 1 | 0% | 1,717 | 1,323 | -23% | 0 | 0 | — |
case-16 | fail→pass | 10,094 | 4,092 | -59% | 1 | 1 | 0% | 1,670 | 1,185 | -29% | 0 | 0 | — |
case-17 | fail→pass | 6,034 | 3,589 | -41% | 1 | 1 | 0% | 907 | 1,110 | +22% | 0 | 0 | — |
case-18 | fail→pass | 1,838 | 4,483 | +144% | 1 | 1 | 0% | 258 | 1,277 | +395% | 0 | 0 | — |
case-19 | fail→pass | 10,682 | 4,646 | -57% | 1 | 1 | 0% | 1,663 | 1,242 | -25% | 0 | 0 | — |
case-20 | fail→pass | 11,694 | 5,129 | -56% | 1 | 1 | 0% | 1,737 | 1,392 | -20% | 0 | 0 | — |
case-21 | pass→pass | 8,917 | 2,676 | -70% | 1 | 1 | 0% | 1,457 | 949 | -35% | 0 | 0 | — |
case-22 | fail→pass | 7,933 | 1,372 | -83% | 1 | 1 | 0% | 1,235 | 650 | -47% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +73 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.